actual_loss = zero_one_loss(node.threshold, node, X, labels)
if actual_loss < error_best:
error_best = actual_loss
best_tresh = -thresh
best_feature = j
i += 1
return best_tresh, best_feature
best = np.inf
best_f = 0
best_t = 0
node = TreeNode(0, 0, None, None, None, None, 0, 0, 0, None)
for j in range(len(X[0])):
node.feature = j
node.threshold = 1
#opt_t = gradient_descend(1, node, X, labels)
opt_t = minimize(quadratically_loss,
0,
args=(node, X, labels),
method='BFGS',
tol=1e-04).x
print("ottimizzo feature: ", j)
err = quadratically_loss(opt_t, node, X, labels)
if err < best:
best = err
best_f = j
best_t = opt_t
